scripts/remote_tts_smoke.py

"""Remote smoke test for Chatterbox Praxy TTS.

This intentionally avoids ASR and LLM calls so we can prove GPU/TTS viability
on short-lived rented instances before running the full S2S path.
"""

from __future__ import annotations

import json
from dataclasses import asdict
from pathlib import Path

from urdu_s2s.schemas import BridgeResult, ReplyResult, SpeechToSpeechRequest
from urdu_s2s.tts_providers import ChatterboxPraxyTTSProvider


def main() -> None:
    request = SpeechToSpeechRequest(
        request_id="tts_smoke_001",
        audio_path=Path("data/raw/benchmarks/gemini_urdu_s2s_v1/audio/bench_001.wav"),
        metadata={"purpose": "remote_tts_smoke"},
    )
    reply = ReplyResult(
        text_urdu="Theek hai, main check kar deti hoon.",
        provider="smoke",
        model="manual",
    )
    bridge = BridgeResult(
        text_devanagari="ठीक है, मैं चेक कर देती हूँ।",
        provider="manual",
        model="manual",
    )
    provider = ChatterboxPraxyTTSProvider(
        output_audio_path=Path("reports/evals/tts_smoke_001.wav"),
        voice_prompt_audio_path=Path(
            "data/processed/voice_anchors/chatterbox_praxy_v1/bench_025.wav"
        ),
        device="cuda",
    )
    result = provider.synthesize(request, reply, bridge)
    print(json.dumps(asdict(result), ensure_ascii=False, default=str, indent=2))


if __name__ == "__main__":
    main()